Awesome

Reconstruct before Query: Continual Missing Modality Learning with Decomposed Prompt Collaboration

This repository contains the code release of RebQ, from our paper:

Reconstruct before Query: Continual Missing Modality Learning with Decomposed Prompt Collaboration

Shu Zhao, Xiaohan Zou, Tan Yu, Huijuan Xu

Pennsylvania State University, NVIDIA

arXiv:2403.11373, 2024.

If this code and/or paper is useful in your research, please cite:

@article{zhao2024rebq,
  title={Reconstruct before Query: Continual Missing Modality Learning with Decomposed Prompt Collaboration},
  author={Shu Zhao and
          Xiaohan Zou and
          Tan Yu and
          Huijuan Xu},
  journal={arXiv preprint arXiv:2403.11373},
  year={2024}
}

Installing Dependencies

We tested our code on Ubuntu 22.04 with PyTorch 1.13. You can use environment.yml and requirements.txt to install dependencies.

Data Preparation

Download UPMC-Food101 and MM-IMDb datasets according to the MAP repo and organize them as following:

data
├── MM-IMDB-CMML
│   ├── images
│   ├── labels
│   └── MM-IMDB-CMML.json
└── UPMC-Food101-CMML
    ├── images
    ├── texts
    └── UPMC-Food101-CMML.json

Run

bash scripts/food101_both_0.7.sh

Acknowledgement

1. Missing-Aware Prompt
2. A Pre-trained Model-based Continual Learning ToolBox